Scaffold, implement, and test new GAIA agents as Python Agent subclasses with @tool methods.
npx skills add https://github.com/amd/gaia --skill gaia-build-agent-amd
Please help me install this Agent Skill. Skill: gaia-build-agent Source: https://github.com/amd/gaia/tree/main/.claude/skills/gaia-build-agent Command: npx skills add https://github.com/amd/gaia --skill gaia-build-agent-amd
gaia agent init to generate a starter package, with a --layout hub option for publishable hub packages. - Agent Implementation Guidance: Covers writing the Agent subclass, registering @tool functions with model-facing docstrings, composing KNOWN_TOOLS mixins (rag, file_io, shell, browser, etc.), and implementing _get_system_prompt(). - Testing and Eval Discipline: Enforces running the real CLI, unit-testing tool logic with a mocked LLM, and running gaia eval agent against baselines whenever LLM-affecting surfaces change. - Use Case: You want to add a new in-core agent to the GAIA repo. The Skill walks you from gaia agent init my-agent through registry wiring, real CLI testing, and eval comparison, then hands off to the agent-hub-release skill for publishing. ## Quick Start Ask the AI to build a new GAIA agent named my-agent, scaffold it with the SDK, register its tools, and test it through the real CLI.High-intent search queries and answers about installing and using this skill.
Run `gaia agent init my-agent` to scaffold a starter package, then write a class inheriting from the base Agent in src/gaia/agents/base/agent.py. Register capabilities as @tool-decorated methods and implement _get_system_prompt() to define the agent's behavior.
Decorate methods with @tool; the docstring becomes the schema the model sees, so write one line of intent plus each argument. Before writing new logic, check KNOWN_TOOLS in src/gaia/agents/registry.py and compose existing mixins like rag, file_io, shell, or browser.
Leave model_id unset so the agent inherits the base default, Gemma-4-E4B-it-GGUF. Sharing one model across agents keeps a single model resident and avoids eviction and cold reloads when switching agents. Only override it when the agent genuinely needs a different model.
In-core agents are added to src/gaia/agents/registry.py and optionally get a CLI subparser. Hub packages are auto-discovered from their gaia-agent.yaml manifest, so no registry edit is needed as long as python.entry_module and entry_class point at your class.
Run it whenever you write or change LLM-affecting surfaces: system prompts, tool docstrings, tool schemas, the model, or error classification. Compare results against the committed baseline, since unit tests with a mocked LLM do not catch LLM behavior regressions.
Do not use it for tuning an existing agent's prompt, adding a single tool to an existing class, or releasing an already-built agent. Those cases belong to prompt-engineering work, general Python development, or the agent-hub-release skill respectively.
OgenticAIInstalls agents, skills, hooks, and CLAUDE.md templates into a new repository.
scouzi1966Build and publish a PyPI wheel from an existing AFM binary.
onehungsolow303-pngMap Gaia subsystems and data flow to diagnose batchmode bake issues.
ZeliperDelegate coding tasks to a coder-agent via the Task tool.